Why Election Polls Keep Getting It Wrong

Election after election, pre-vote surveys miss the final result by margins that make front pages and fuel conspiracy theories. Yet polling remains the best systematic tool journalists and campaigns have for gauging public opinion. So what goes wrong, and is it getting worse?

The short answer is that surveying human opinion is genuinely hard. Classical polling relied on random telephone sampling, call enough people at random and, statistically, you get a representative picture. But response rates have collapsed over recent decades as caller ID became universal and mobile phones replaced landlines. Pollsters now work with samples that are far less random than they once were, which introduces what statisticians call selection bias.

The Weighting Problem

To compensate, modern pollsters weight their data. If their sample has too many university graduates relative to the general population, they mathematically downweight those responses. This sounds sensible, but weighting is an art as much as a science. Choosing which variables to weight, age, education, geography, past voting behaviour, involves judgement calls, and different firms make different calls, which is why polls on the same race sometimes diverge sharply.

There is also the problem of "herding," where polling firms, conscious of reputational risk, adjust their results toward the emerging consensus rather than publish an outlier. This can create a false impression of certainty right before an upset result.

What Polls Are Actually Good For

None of this means polls are useless. Aggregating many polls together reduces the noise from any single firm's methodological choices, which is why polling averages are generally more reliable than individual surveys. Polls are also more informative about long-term trends and relative shifts in opinion than about pinpoint final outcomes.

The real problem may be how polls are reported. A survey showing one candidate six points ahead with a margin of error of four points is not evidence of a comfortable lead, it is evidence of a genuinely competitive race. When media coverage treats every poll as a precise prediction rather than a probabilistic snapshot, public trust erodes the moment reality diverges. Better literacy about what a poll can and cannot tell us would serve everyone well.